Nm

Ja

b

c

a

KNMIHI

1

ceosfl

tFbZWrUWgWa

n

h0

Social Networks 41 (2015) 1–17

Contents lists available at ScienceDirect

Social Networks

jo ur nal homepage: www.elsev ier .com/ locate /socnet

etwork effects on organizational decision-making: Blended socialechanisms and IPO withdrawal�

ason Owen-Smitha,∗, Natalie C. Cotton-Nesslerb, Helena Buhrc

University of Michigan, Department of Sociology, #3001 500 S. State St., Ann Arbor, MI 48109-1382, United StatesBentley University, Department of Management, 175 Forest Street, Waltham, MA 02452, United StatesThe Northwestern Institute on Complex Systems (NICO), 600 Foster Street, Evanston, IL 60208-4057, United States

r t i c l e i n f o

eywords:etwork mechanismultiple networks

nter-organizational networksigh technology industries

nitial public offering

a b s t r a c t

This paper develops a new approach to the study of network effects in organizations and markets byproposing that structural influences on social and economic action result from contingent blends ofwell-understood social mechanisms. We emphasize the interplay of three different network processes:resource and information transfer, status signaling and certification, and social influence. Different mixesof these mechanisms characterize disparate networks because the obligations imposed by ties and thecapacities of partners result in situations where mechanisms amplify or diminish one another. We testhypotheses about mechanism interactions using four years (1997–2000) of data on high-technology IPOsthat situate organizational decisions about whether to withdraw an offering in two distinct networks. We

find that network mechanisms exert multiple moderating effects on one another and that those effectsvary systematically across venture capital syndicate and director interlock networks. These findings helpto explain why different networks exert disparate effects, why the effects of some structures change astheir larger contexts shift, and why even very successful organizations can sometimes find themselveshamstrung by their connections.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

Networks shape social and economic action through multiple,o-occurring mechanisms (Podolny and Baron, 1997; Fernandezt al., 2000; Podolny, 2001). We develop the idea that social and

rganizational connections blend together at least these threeocial mechanisms: Among other things, relationships channelows of tangible and intangible resources (Granovetter, 1973; Burt,

� This paper was a fully collaborative effort, all three authors made essential con-ributions to the work. We acknowledge research support from the National Scienceoundation (grant # 0545634) and the Alfred P. Sloan Foundation. Our efforts haveenefited from able research assistance from Jessica Galoforo, Tammy Yang, Mikehu, Erica Jankelovitz, David Weisman, Jeff Lundy, Grace Kang, and Josh Cykiert.e appreciate useful commentary from members of the US Knowledge Economy

esearch team at the University of Michigan and from audience-members at theniversity of Arizona’s McGuire Entrepreneurship Center, the Economic Sociologyorkshop and the 2008 Entrepreneurship Conference at the University of Michi-

an, as well as the 2008 Sunbelt and American Sociological Association Meetings.e also thank Jerry Davis for wise counsel. Any remaining mistakes or infelicities

re ours alone.∗ Corresponding author. Tel.: +1 734 976 0700.

E-mail addresses: jdos@umich.edu (J. Owen-Smith),cottonnessler@bentley.edu (N.C. Cotton-Nessler), hbuhr@umich.edu (H. Buhr).

ttp://dx.doi.org/10.1016/j.socnet.2014.11.004378-8733/© 2014 Elsevier B.V. All rights reserved.

1992), signal status and membership (Podolny, 1993; Zuckerman,1999), and convey influence (Friedkin and Johnsen, 1990). Weseek to explain how these mechanisms intensify or diminish oneanother’s effects to enforce tradeoffs on or offer unanticipated gainsto the participants in particular networks.

We begin with the observation that relationships in a networkare not static indicators of a single social process. While it is tempt-ing to assume, for instance, that the fact of a connection entails thetransfer of valuable information between partners, it need not. Net-work ties are clean representations of messy interactions. They thusstand in for complicated social processes that can shape action andoutcomes through several means. We contend that network scho-lars must begin to elaborate the ways in which structures reflect theworkings of multiple social mechanisms that might act in concert orat cross purposes, depending on the actors, activities, and context

that characterize a particular empirical network. At least three pro-cesses – information transfer,1 status signaling, and social influence– are muddled up in most relationships. Moreover, we expect the

1 For the sake of rhetorical simplicity we use the phrase ‘information transfer’ toencompass all movements of valuable tangible and intangible resources throughnetwork connections.

2 cial Ne

st

metnete

ioocumptpo

avpiwltt

ttsurhFit(pwcas

twsisdoaatatwamupct

There are three reasons that decisions about whether to with-draw an IPO and forgo other sources of liquidity offer a particularlyrigorous test of our approach to network effects on organizational

2 One reason firms commonly withdraw an IPO is to enable them to pursue anattractive financial alternative such as a merger or acquisition. (Despite the fact thatwhile there is a valuation benefit to companies which are acquired during the IPO

J. Owen-Smith et al. / So

trength and salience of each mechanism to vary across differentypes of activities and partners.

Put succinctly, this means that network theorizing must becomeore contingent as the mechanisms at work and thus observable

ffects of social structure will vary from context to context. We seeko contribute to the development of a more contextual theory ofetwork multiplicity that expands our ability to explain the variedffects exerted by differently configured networks by emphasizinghe blends of mechanisms that characterize their workings. To thatnd we make two related arguments.

First, we contend that an individual or organization’s positionn a given network activates several processes that condition eachther’s effects (cf. Smith et al., 2012). Networks influence action andutcomes through blends of mechanisms. Second, we argue that theontent of ties and characteristics of their participants matter fornderstanding how mechanisms blend. All networks contain someix of information transfer, status signaling, and social influence

rocesses. However, characteristics of the parties involved and theypes of relationships linking them affect how these simultaneousrocesses operate – in tandem or at cross-purposes – to producebservable outcomes.

We test hypotheses about when and how different social mech-nisms amplify or diminish each others’ effects using data on theenture capital syndicate and director interlock networks of theopulation of high-technology companies that announced their

ntention to go public between 1997 and 2000. We seek to explainhy corporations that have signaled their intention to pursue a

iquidity event by filing a prospectus indicating they will under-ake an initial public offering (IPO) change course by withdrawingheir IPO to remain private and independent.

Research in finance and entrepreneurship has demonstratedhat the decision to withdraw an IPO has to do with a corpora-ion’s (or its advisors’) perceptions of the firm’s chances for marketuccess. The financial characteristics of offerings and companiesnderpin common firm level explanations for withdrawal. Greaterevenues, less debt, and larger offerings all decrease the likeli-ood of withdrawal (Busaba et al., 2001; Zhao, 2005; Dunbar andoerster, 2008). In addition, a corporation’s choice of partners ismportant: underwriter prestige and the choice of whether or noto have venture capital backing affect the likelihood of withdrawalBoeh and Southam, 2011). We extend this insight into the role ofartners and propose that key networks influence the likelihood ofithdrawal. The decision to forgo a desired outcome stems from a

ombination of resource and information transfer, status signaling,nd social influence mechanisms that vary across the interlock andyndicate networks.

Our argument for a more contingent, contextualized approacho network theorizing creates new challenges for empirical net-ork studies, including this one. Our conclusion and implications

ection more fully discusses the general characteristics of a behav-oral network theory that takes the kinds of contingencies we studyeriously. For now it is important to highlight a key tension: theesire to create generalizable theory about network processes isften at odds with the goal of contextualizing our understandingbout those same processes. On the one hand, we seek to articulatend test hypotheses that are general enough to provide purchase onhe effects networks exert in many settings. On the other, we arecutely conscious that our core argument depends on the notionhat there are likely to be few universal relationships between net-ork measures and outcomes. In an effort to make claims that are

s generalizable as possible while still doing justice to our argu-ents about the contextual dependence of networks, we take the

nconventional step of introducing our empirical setting beforeresenting our theoretical arguments in order to allow the empiri-al details of our focal setting and actors to inform the analytic workhat follows.

tworks 41 (2015) 1–17

2. Setting

We develop and test our hypotheses in an examination of firms’decisions about whether or not to realize an initial public offer-ing (IPO) in high technology industries in the period from 1997to 2000. The IPO is a bellwether event in the life of a young firm(Gompers and Lerner, 1999) and is a routinely used outcome mea-sure in finance (Lerner, 1994; Ritter and Welch, 2002) and strategy(Pollock and Rindova, 2003; Higgins and Gulati, 2006). Yet firmsthat indicate their desire to go public need not complete the pro-cess. At any time after announcing intentions to IPO, a firm may“withdraw,” canceling its ability to offer securities on the publicmarket.

The decision to pursue an IPO carries the considerable costs ofdeveloping and filing a prospectus, establishing underwriters, andworking to generate demand for an issue, a process called “bookbuilding” (Busaba, 2006). When an organization files an S-1 formwith the Securities and Exchange Commission (SEC) it is publiclyannouncing a strong preference to realize its IPO. In addition to thesunk costs of beginning the IPO process, companies and managershave disincentives to forgo an IPO.

Withdrawing an IPO diminishes the reputation of both the orga-nization and the individuals that run it (Acimovic and Lyn, 2000).Companies that withdraw IPOs are much less likely to make a sec-ond, successful attempt to go public (Dunbar and Foerster, 2008),and if they do, they suffer a valuation penalty as a result of beingperceived as a higher risk (Lian and Wang, 2009). Thus firms thatstay independent in the wake of a withdrawn offering have to relyon other – frequently more expensive – sources of financing. With-drawing an IPO without an alternative liquidity event also defersinsiders’ and financiers’ ability to profit from pre-IPO equity andoptions.2

Nevertheless, IPO withdrawals occur fairly often. Over a 20 yearperiod, Lian and Wang (2012) estimate that slightly more than20% of all attempts to go public result in a withdrawn registration.Companies in the IPO pipeline can withdraw for numerous rea-sons including negative (bankruptcy) and more positive (appealingacquisition opportunities) outcomes. However, some percentage ofcompanies step away from equity markets only to remain alive andindependent or to try the IPO process again at a later date.

We take the decision to withdraw an IPO by a company thatdoes not subsequently go bankrupt or cease to be independentto represent an instance where an organization changes courseto pursue a costly alternative other than the one for which it hasalready signaled a strong preference. In addition to firm level fac-tors (small offerings, few or no revenues) that have been shown toinfluence withdrawal rates (Busaba et al., 2001; Zhao, 2005; Dunbarand Foerster, 2008), we propose that firms’ networks will influ-ence the likelihood of withdrawal. This is consonant with Boeh andSoutham’s (2011) finding that features of a firm’s coalition of back-ers and advisors affect the likelihood of withdrawal. For example,the more prestigious the underwriter and the more active the ven-ture capital investors, the lower the likelihood a firm will withdraw(Boeh and Southam, 2011).

process, this benefit exists only if the company is acquired before formally with-drawing the IPO (Lian and Wang, 2009.) By focusing our discussion and analyses onfirms that withdraw and remain independent, we avoid the possibility that similarfinancial incentives may drive both the decision to file for an IPO and the decisionto withdraw.

cial Ne

aocatfithoycttaopo

3

3

dabtraaoemgntltotraeoii

aiscitidaoi“tg

oto

J. Owen-Smith et al. / So

ction. First, filing an S-1 is a clear, costly, and public declarationf an organization’s intention to pursue a particular, observableourse of action. Second, withdrawing the IPO represents an equiv-lently public, costly, and observable declaration of the intentiono shift course and pursue an alternative. Third, the period beforeling an S-1 is one of the very few times that a private firm, subjecto many fewer regulatory constraints than its public counterparts,as much more leeway to strategically configure key networks inrder to maximize the chances of achieving its goal. Thus, our anal-sis of withdrawal decisions as a function of position in ventureapital syndicate and director interlock networks at this time inhe organizational life course represents a very hard case for ourheory because the interests of most relevant actors are likely to beligned in favor of pursing liquidity. As a result the seemingly curi-us choice to pursue an IPO only to later withdraw from the processrovides a substantive motivation for our theoretical explorationf blended network mechanisms.

. Theory in context

.1. Three mechanisms: pipes, prisms, and peers

In order to explain how organizational positions in these twoifferent networks combine to influence the decision to withdrawn IPO, we turn to three well-known network mechanisms. Weegin with Joel Podolny’s (2001) memorable metaphorical distinc-ion between network pipes and prisms. When understood as pipes,elationships work by channeling resources from place to place in

differentiated social structure. Those resources can be tangible,s in the case of financial investments made by venture capitalists,r intangible, as when a firm’s partners bring them timely and rel-vant information about competitors, markets or other importantatters of business. Regardless of whether the resources are tan-

ible, this mechanism proposes that valuable things move throughetworks. More central positions thus offer greater opportuni-ies and advantages. Network theories that emphasize individualevel differences in social capital (Lin, 2001; Burt, 2005) rest onhis conception. When viewed as prisms, in contrast, networksrder the experience of complicated settings by signaling the rela-ive status of participants to important outsiders who control keyesources (Podolny et al., 1996). Network theories that treat tiess means to certify uncertain quality (Stuart et al., 1999) or tostablish membership in a valued category (Zuckerman, 1999) restn this understanding. Here status benefits can accrue to central-ty in networks whether or not valuable resources such as timelynformation pass through them.

We also incorporate a third, somewhat less strategic mech-nism that leads networks to shape economic outcomes: socialnfluence. Research in this tradition “. . .links the structure ofocial relations to the attitudes and behaviors of the actors whoompose a network” (Marsden and Friedkin, 1993: 127). Socialnfluence emphasizes the shared norms (Useem, 1984), expecta-ions (Coleman et al., 1966), mindsets (Galaskiewicz, 1985), anddentities (Podolny and Baron, 1997) that pressure egos to makeecisions based on the perspectives, actions and experiences oflters. This mechanism highlights conformity and social pressuresn decision-making and may be independent of or even antithet-cal to ego’s strategic goals. In keeping with Podolny’s “pipes” andprisms,” we dub the social influence mechanism “peers” to denotehe role that relationships play in shaping individual attitudes androup norms (Friedkin, 2003).

More abstractly, we propose that the pipes mechanism dependsn timely flow of potentially valuable information and resourceso ego through their partners. The prisms mechanism dependsn signals of quality inferred by third parties from the degree of

tworks 41 (2015) 1–17 3

visibility of ego’s more or less well-connected partners. The peersmechanism depends on the accessibility and salience of a partner’sopinions or expectations to ego. As we mobilize them here boththe pipes and peers mechanisms emphasize flows of informationthrough networks, but only the former requires that the informa-tion be of potential value to ego. The prisms mechanism does notrequire any flow of information through ties. Instead, others’ infer-ences based on observing their presence drive outcomes.

How these three mechanisms blend in particular empirical sett-ings, we contend, is deeply dependent on the quality of attentionthat partners can pay to ego and the factors which make partners’opinions and knowledge salient to ego’s decisions. The key empir-ical questions for predicting how these networks might blend inparticular contexts are: (1) does increased centrality come at thecost of attention; and (2) what does the character of attention in anetwork suggest about the sources of particular partners’ salienceto a given ego. These questions, we contend, cannot be answered inthe abstract. In other words, we propose that the quality of atten-tion partners can pay to ego is key to both the value and the salienceof information that might flow through networks. Thus we empha-size partner’s attention as a way to understand when differentmechanisms are more likely to be more or less determinative oforganizational decisions.

3.2. The networks of high-technology pre-IPO firms

At least two different networks might influence the withdrawaldecision. Young technology firms often receive early financial sup-port from VCs. Those investments have important implicationsfor a company’s eventual success (Stuart et al., 1999; Podolny,2001; Gulati and Higgins, 2003) because they influence stock mar-ket investors’ valuations of potential offerings and offer access toexpertise in addition to capital. VC investments have two furtherfeatures that make them interesting for our purposes.

First, VCs tend to be “hands on” investors who exert real influ-ence on the strategy and management decisions of their portfoliocompanies (Stross, 2000). Ties between financiers and the compa-nies in which they invest transfer important tangible and intangibleresources to the company (Hsu, 2004). In addition to cash, venturecapitalists bring extensive managerial and sometimes technicalexpertise to the organizations in which they invest. VCs also rou-tinely serve as matchmakers by, among other things, helpingyoung companies identify and hire the executive officers who willshepherd them through the IPO and subsequent growth. Second,contemporary VCs rarely invest alone. Instead they syndicate dealswith other firms in the industry. Thus it is possible to position port-folio companies and their investors in a syndication network thatspans the private equity market (Hochberg et al., 2007; Kogut et al.,2007).

Pre-IPO startups also maintain important ties to already publiclytraded corporations. Those connections often take the form of boardof director interlocks (Mizruchi, 1996; Useem, 1982; Mizruchi,1992; Palmer et al., 1995). Interlock networks are configured differ-ently than VC financing networks. Here, a technology firm (ego) hasa tie through a shared director to another firm (partner) whose sim-ilar ties to other companies make it more or less visible to outsiderswho use the director interlock to make inferences about the pre-IPO company’s quality. Both director interlocks and VC investmentscan convey status and allow the transfer of information. Only VCinvestments also come with direct transfers of material resourcesto the firm. But when entrepreneurs have a choice of venture cap-

italists they often accept lower levels of financial investment inreturn for access to the attention and expertise of established VCpartners (Hsu, 2004). We thus frame most of the discussion thatfollows in terms of information that flows through networks while

4 cial Ne

rr

3

cva(bwffl

wgtatanatepstplh

Hbo

3

nappoptemnma

VvvsafIta‘o

Hnt

mechanism when more and better flows of information reach it viaits direct connections to partners. Earlier we argued that the qual-ity of attention that partners can pay to ego conditions the qualityof information that they can bring to ego’s deliberations. Whether

3 The efficacy of network pipelines depends on timely movement of potentiallyvaluable tangible and intangible resources from partners that give one a leg up incompetitive environments. In contrast, the network peers mechanism depends onlyon flows of information that make the attitudes and decisions of others clear andsalient (Friedkin, 1993). While nuanced, the difference is an important one. Take, forexample, the oft-noted tendency of cohesive networks, where most participants areinterconnected, to generate redundant information (Granovetter, 1973; Burt, 1992)groupthink (Coleman et al., 1966; Lave and Wenger, 1996) or a similar mindset(Galaskiewicz, 1985). In this view partners shape the grounds ego uses to make

J. Owen-Smith et al. / So

ecognizing that many other things are also transferred throughelationships.

.2.1. Access, search, and network pipesThe key insight underlying the networks as pipes view is that the

ollective structure created by relationships in a field can yield indi-idual benefits for some. The key imagery here is one of increasedbility to compete that results from a salutary position. As Burt2005: 4) notes: “One’s position in the structure of . . .exchanges cane an asset in its own right.” If we take the metaphor of the net-ork pipeline seriously, then one’s competitive advantages stem

rom beneficial levels of access to information or resources thatow from partners.

All other things equal, then, we expect that technology firmshose investors and directors are better connected will have

reater access to timely and potentially valuable informationhrough those connections. However, we also contend that thebility of investors and directors to convey what they know to par-icular firms in a timely and effective manner depends on theirbility to attend more exclusively to the companies they are con-ected to. Firms that benefit from the attention of well connectedlters will thus have an advantage by virtue of being better ableo select the appropriate time to go public, to secure more ben-ficial backing in the form of high status underwriters, and toresent themselves more effectively in the often arduous ‘roadhow’ presentations that precede an IPO. In other words, whenhese networks are conceptualized as resource pipelines, we expectre-IPO firms that are more central to be better advised and more

ikely to succeed in their pursuit of a successful IPO. We thusypothesize:

1. Ceteris paribus, access to information and resources throughoth VC and interlock network ‘pipes’ will decrease the likelihoodf IPO withdrawal.

.2.2. Certification, status, and network prismsNetwork ties also order outsiders’ experiences of social and eco-

omic settings. Structures linking participants in a field provide social map that outsiders utilize to evaluate a given partici-ant’s prospects. In settings where the true value of organizations,roducts, or services is uncertain, investors and customers minebserved relationships for clues about quality (Podolny, 1993). Theerformance of initial public offerings (IPOs) is thus conditioned byhe presence or absence of firms’ ties to prominent partners (Stuartt al., 1999). Prism effects help participants by making them seemore valuable than competitors who lack the right kinds of con-

ections. In this view, the attention and attributions of outsidersay be largely independent of the intentions and needs of the egos

nd partners that are party to a tie.Strong prism effects created by investments from high status

Cs and board membership of prominent directors increase theisibility of pre-IPO corporations to investors. Such ties also signalalue in uncertain IPO markets, increasing the likelihood that out-ide investors will be willing to purchase shares in a new offeringnd the price of those shares. The status hierarchy of these fields isairly transparent (Podolny, 1993; Hsu, 2004). We thus expect pre-PO firms that have not secured high profile partners will perceiveheir chances of success to be lessened and as a result to withdraw at

higher rate. Thus we predict that pre-IPO companies with strongprism’ networks to have and know they have a greater likelihoodf IPO success. Thus:

2. Ceteris Paribus, signals of value from ties to prominent part-ers through both VC and interlock network ‘prisms’ will decreasehe likelihood of IPO withdrawal.

tworks 41 (2015) 1–17

3.2.3. Influence and the salience of network partnersSocial influence has been little studied in the context of IPOs.

The peers mechanism focuses on the expectations, obligations, andsocial pressures that accompany relationships. The pipes and peersmechanisms can be difficult to distinguish analytically becauseboth rely, though in different fashions, on the idea that informationflows through networks (Burt, 1992; Marsden and Friedkin, 1993).3

For the peers mechanism, partners shape the grounds ego usesto make decisions. Partners’ knowledge and expectations, whichare often garnered through their connections to others, can ren-der some types of actions illegitimate for ego while introducingor legitimating alternatives. The information partners convey mayor may not increase ego’s competitive advantage and the tie thatchannels that information may or may not be positively perceivedby outsiders. What flows through ties in the peers conceptualiza-tion is not resources or signals but social pressure and expectations(Uzzi, 1996). In short, the network peers mechanism treats rela-tionships as means for partners’ opinions and experiences to shapeego’s conception of a situation.

A key aspect of influence involves the partner’s possessionof some source of expertise that might influence ego’s decision,combined with structural arrangements that make that experi-ence visible and salient to ego. In particular, we contend thatexperience with other entities in the network gives the partner aparticular source of influence. When partners have experience withalternatives, their connections to ego make those alternatives vis-ible. Ego, in turn, is more aware of different options and thus morelikely to pursue a new course.

To put things more concretely, we expect that network ties topowerful partners who have been investors or directors for anotherfirm that withdrew its IPO will make the possibility and poten-tial legitimacy of withdrawal more salient to ego. The increasedsalience of withdrawal will in turn make that option a more likelyone for ego. We thus propose:

H3. Ceteris paribus, connections to VC and interlock network‘peers’ that have experience with a prior withdrawal will increasethe likelihood of IPO withdrawal.

3.3. Blending mechanisms

3.3.1. Prisms moderate pipesEgo’s relationships signal greater status (prism effect) when

its partners are better connected to others in the network. Theroute to competitive advantage through the prisms effect for ego isto have partners whose own expansive connections make them(and by extension, ego) more prominently visible to observers.In contrast, ego’s relationships provide advantages via the pipes

decisions. Partners’ knowledge and expectations, which are often garnered throughtheir connections to others, can render some types of actions illegitimate for egowhile introducing or legitimating alternatives. The information partners convey mayor may not increase ego’s competitive advantage and the tie that channels thatinformation may or may not be positively perceived by outsiders.

cial Ne

pnotemaevn1

memwfogee

dtjuvtacttaUt

H‘l

tattnmocmt2ct

sppa1rI

pt

We also suggest, though we can find no published examplesto support the idea, that ego’s recognition of partners’ status con-tribution also creates dependence. Small businesses who land a

5 In contrast to H4, a positive moderation is hypothesized here. However, since thedirect effects are hypothesized to be negative, the interaction term is also expectedto be negative.

6 In the terms French and Raven (1959) develop, we are arguing that a partner’sability to punish ego makes its experiences more salient in ego’s decision-making.

J. Owen-Smith et al. / So

risms amplify or interfere with pipe effects thus depends on theetwork being examined. When ties impose continuing obligationsn partners who have limited capacities to manage and maintainheir connections, having ties to prominent partners who increasego’s status can diminish the flow of valuable resources and infor-ation to ego. Put simply, when partners who have lots of ties (and

re thus more prominent) cannot attend as carefully to a particulargo, the quality of information they transfer will decline. Indeed thisery dynamic may account for findings that suggest having promi-ent directors adds very little to corporate performance (Mizruchi,996; Davis and Robbins, 2005).

Thus we argue that director interlock networks – where infor-ation that flows from partners (other corporations with whom

go shares a director) depend on individuals for whom added boardemberships multiply the information, meetings, and concerns tohich they must attend – are just this kind of network. In order

or ego to benefit from information made available through ties tother firms in this type of network, its directors must be able tolean relevant tidbits from others with whom they share a differ-nt board room, recognize the importance of that knowledge togo, and convey it in a timely fashion.

Ego’s status in a director interlock network increases because itsirectors sit on more and more prominent boards for other corpora-ions. Thus, increases in status mean that ego’s directors hold moreobs (appointments as directors of other corporations), which takep the directors’ time and focus. As a result, ego’s ability to gainaluable information though highly connected directors rests onhe ever more overloaded mechanism of individual cognition andttention (March and Simon, 1958). The basic idea here is that theharacteristics (lots of ties to other partners) that allow directorso increase ego’s status diminish those directors’ ability to attendo ego because of the varied and competing calls on their timesnd the sheer volume of information to which they are exposed.nder such circumstances, increased status signaling will diminish

he effect of information transfer. Thus we expect:

4. Ceteris Paribus, as signals of value from interlock networkprisms’ increase, the effect of access to information through inter-ock network ‘pipelines’ will weaken (become less negative).4

In contrast, when partners’ capacities to manage and main-ain connections are enhanced, having prominent alters will alsomplify the flow of valuable information to ego. VC contributionso the firms where they invest are likely to increase as the sta-us of the financier grows because visibility in the syndicationetwork accompanies experience derived from engagement withore deals and partners. The internal, partnership-based structure

f most VC firms makes the advice given to particular portfolioompanies the province of individual general partners, but com-on partnership meetings ensure that the collective experience of

he firm is brought to bear in that advice (Stross, 2000; Perkins,007). The standard VC business model yields a mix of exclusive,ustomized attention and network reach to portfolio companieshat are the targets of investment from well-connected VCs.

The standard organizational practices of VC firms – staged andyndicated investments linked to particular milestones and accom-anied by active managerial engagement by an individual generalartner – ensure that investors have a great degree of oversightnd authority over their portfolio companies (Gompers and Lerner,

999; Lerner, 2012). That attention is most strongly felt when expe-ienced and higher status investors exert significant input in thePO decision (Lerner, 1994). We argue that the business model of

4 A negative moderation is hypothesized here. However, since the direct effects ofipes and prisms are hypothesized to be negative, the interaction term is expectedo be positive.

tworks 41 (2015) 1–17 5

contemporary VC firms makes the syndication network a contextwhere an increasingly strong prisms mechanism will amplify theeffect of the pipes mechanism. Thus, we expect:

H5. Ceteris Paribus, as signals of value from VC network ‘prisms’increase, the effect of access to information through VC network‘pipelines’ will strengthen (become more negative).5

In concrete terms, our argument in this section suggests thatthere are some networks where status signals and social capitalamplify one another, offering cumulative advantages to egos whoare able to secure ties to well-connected alters that are able toattend closely to them. In other structures, however, prisms willinterfere with pipes, forcing egos to tradeoff between access tovaluable information and signals of quality. The key to this dif-ference is the capacity of partners to attend to the needs andcircumstances of ego.

3.3.2. Mixing pipes and prisms with peersThree conditions are necessary for social influence to shape a

firm’s decision to forgo an IPO after filing to pursue one. First, part-ners must have experience with a withdrawn IPO. Second, a tielinking ego to the partner makes that experience visible to ego byallowing information about prior withdrawal to transfer. Third, theadvantages that accrue to ego through its partners must make thatvisible experience salient in ego’s decision-making. Many thingscould focus ego’s attention, but we emphasize cases where alterhas power over ego because of ego’s dependence on alter (Pfefferand Salancik, 1978).6 Networks create such power differentials bymaking ego more or less reliant on particular partners for benefitsderived from relationships (Blau, 1964). That reliance is most obvi-ous when the partner can remove something of benefit to ego. Inthe networks we study, partners can change ego’s ability to com-pete by (a) interrupting the flow of tangible or intangible resourcesto ego; or by (b) severing its ties to ego thus removing a valuablestatus signal.

In other words, either information transfer or certification canamplify the effects of influence by making a partner’s visibleexperience more salient to ego. The question of which direc-tion one predicts is an empirical one that hinges on identifyingwhether pipes and prisms offer substitutable or complementarysources of advantage. For instance, Haunschild and Beckman (1998)demonstrate that when firms have access to alternative sourcesof information through CEO memberships in professional or tradeassociations, board influence on acquisition decisions declines. Inthe terms we use here, this finding means that when ego’s depend-ence on partners decreases, so too does the partners’ influence onego’s decision-making. In other words, pipes positively moderatepeers.

We suspect that other sources of power could accomplish the same goal. Podolnyand Baron (1997), for instance, argue that ego’s identification with alter createssalience. Burt (1987) makes a similar claim about competition. In a study of board ofdirector influence on strategic decisions about acquisitions, McDonald et al. (2008)demonstrate that director experience with similar acquisitions at the other com-panies (partners, in our terms) on whose boards they also serve increase the valueof ego’s acquisitions. The positive influence of the partner’s experiences, however,is magnified by a governance structure that makes boards independent of manage-ment. In other words, formal authority can also make partners’ sources of influencemore salient to ego.

6 cial Ne

wmfscwsap

aifpbe

Hwe

nstapdw

Hii

gtCtomhwaec

mittttd

4

sot

(toa

the SEC. We collected these forms from the EDGAR database andchecked their validity with the SDC New Issues dataset in order todetermine the date when an offering was realized or withdrawn.

8 The firms we examine in this sample are a subset of a larger dataset that tracks

J. Owen-Smith et al. / So

ell-known company as a customer often advertise the achieve-ent, knowing that having a prominent customer is looked upon

avorably by other potential customers. As a result, the relation-hip with the prominent customer may be attended to morearefully and nurtured by the business owners. Losing a single,ell-connected partner can thus deal a significant blow to out-

ider valuations of ego to the extent that the relationship served certifying purpose. In short, prisms too can positively moderateeers.

To summarize, when ties impose continuing requirements forttention on partners and when partners have limited capacities ass the case in the interlock network, we expect increased attentionrom partners to amplify their influence. In other words, when aartner’s attention is a limited resource the value of their contri-ution to ego will be increased as the relationship becomes morexclusive. Thus we expect:

6. Ceteris Paribus, as director attention through interlock net-ork ‘pipelines’ increases, the salience of director withdrawal

xperience grows, increasing the likelihood of withdrawal.

In differently configured networks, however, we expect part-ers with many other connections will be able to send strong statusignals while also increasing ego’s information and resource advan-ages. When information and certification mechanisms amplify onenother, as we expect to be the case in the VC syndicate network, aartner’s visibility will make their experiences more salient in ego’secision-making, thus increasing the likelihood of influence. Thus,e expect:

7. Ceteris Paribus, as signals of value from VC network ‘prisms’ncrease, the salience of investors’ withdrawal experience grows,ncreasing the likelihood of withdrawal.

In concrete terms, these two hypotheses suggest that gainingreater capacity to compete via network pipes or prisms can some-imes make it less likely that ego will succeed in its stated goals.ontra our first hypotheses (H1 and H2), we expect dependencehat results either from partners who can interrupt resource flowsr whose departure will send negative status signals to make itore likely that firms will withdraw their IPOs when those partners

ave prior experience of withdrawal because ego’s dependenceill make that experience more salient. We expect that both pipes

nd prisms sometimes blend with peers to enforce tradeoffs ongo’s ability to pursue its goals using resources garnered from itsonnections.

Taken together, hypotheses 4–7 suggest the beginnings of aechanism-based theory of network multiplicity. Networks vary

n terms of the key resources that flow from partners to egos, inerms of the type and limitations of partners themselves, and inerms of the kinds of signals outsiders might seek to glean fromhe ties they observe.7 These factors lead multiple, mutually con-ingent social processes to blend together in different ways, givingisparate networks their distinct character and effects.

. Data and methods

Testing our hypotheses requires a unique dataset that we con-

tructed from multiple sources. Our empirical analysis focusesn U.S. based VC funded high-technology firms that regis-ered an initial public offering with the Securities and Exchange

7 Because the relevant outsiders for both the networks we observe are the sameanalysts and investors) and because we do not attend (ala Gulati and Higgins, 2003)o contextual effects on those outsider’s key uncertainties, we bracket the questionf how outsider perceptions might shape network blends here, but note that this isnother potentially fertile ground for the style of analysis we propose.

tworks 41 (2015) 1–17

Commission’s (SEC) by filing an S-1 prospectus in the years between1997 and 2000. We define high-technology industries using fourdigit Standard Industrial Codes (SIC).8 The firms’ primary activitiesfall in five sectors: drugs and biotechnology, hardware and semi-conductors, medical and laboratory devices, software and Internet,and analytic services. Corporations that declared bankruptcy afterfiling an S-1 but before realizing their IPO were excluded from thesample, as were companies that withdrew an IPO as a result ofmerger or acquisition.9 In total 453 firms are included in the anal-ysis. The sample is unevenly distributed across the four years weobserve due to the overall growth of technology IPOs across thistime period: 15.23% of sample firms filed S-1s in 1997, 13.02% in1998, 35.98% in 1999 and 35.76% in 2000. Table 1 presents the broadsectors and the SIC codes that comprise them along with a countof IPO registrations. The firms in these sectors represent the coreof the high-technology or knowledge economy. These companiesthus partook of and reacted to the rise and fall of the stock marketthat is associated with the technology bubble that built up duringthe late 1990s and ended abruptly in 2000.

4.1. Identifying the population of pre IPO firms

We began by collecting all S-1 filings that firms in our fiveindustrial sectors filed with the SEC. In SEC’s EDGAR database,which includes all electronically filed forms, the primary industryis missing for a large number of S-1 forms. To avoid missing validtechnology prospectuses because of incomplete SEC index data, weadopted a multi-stage data collection strategy. The first stage drewon two databases, SEC Edgar and Thompson Research, to collectthe names of all firms active in any of the 44 SICs identified inTable 1. These searches were based on S-1 filings as well as 10-Kannual reports and thus included both companies that were pub-licly traded in the time period and those that sought to go public.Having identified all active corporations in these industries we col-lected S-1 forms and hand coded the primary industry the filing firmreported on the first page. We treat the company’s own conceptionof its primary industry as the best indicator of the firms’ activities.We exclude S-1s that do not announce an initial public offering byappeal to SDC Platinum’s new issues database and, where neces-sary, hand coding of individual forms. The final dataset consist of453 VC funded firms that filed an S-1 indicating their intention topursue an IPO in the years 1997–2000 and that did not withdrawtheir IPO due to bankruptcy. Table 1 reports the distribution of thoseorganizations across industries and sectors.

4.2. Dependent variables

A surprising number of companies that file IPO prospectusesnever actually go public. Firms that decide to withdraw an ini-tial public offering indicate that decision by filing a form RW with

that the rise and evolution of knowledge intensive U.S. industries. For the purposesof this study, we define a 4 digit SIC to be research or knowledge intensive if R&Dspending by firms whose primary industrial classification is in that sector accountfor 10% or more of the industrial R&D spending in the 3 digit SIC code in which it isnested.

9 Some firms withdraw from the IPO process to obtain an alternate liquidity event– to merge with or get acquired by another firm. If the goal of an IPO is liquidity, thena withdrawal to pursue an alternative source of liquidity may not truly be a decisionto forego a desired outcome. As a result, we exclude firms that are merged/acquiredwithin four years of the IPO withdrawal. Similarly, firms that are failing and declarebankruptcy are not so much choosing to forego a desired outcome, but rather areforced to leave the IPO process despite their hopes or intentions.

J. Owen-Smith et al. / Social Networks 41 (2015) 1–17 7

Table 1Sample firms by industry and sector.

SIC Industry Sector IPO reg.

2833 Medicinal chemicals and botanical products Drugs/biotech 12834 Pharmaceutical preparations Drugs/biotech 272835 In vitro and in vivo diagnostic substances Drugs/biotech 12836 Biological products, except diagnostic substance Drugs/biotech 123571 Electronic computers Hardware 33572 Computer storage devices Hardware 13575 Computer terminals Hardware 03577 Computer peripheral equipment, NEC Hardware 83578 Calculating and accounting machines, except electronic comp. Hardware 13661 Telephone and telegraph apparatus Hardware 213663 Radio and television broadcasting and communication Hardware 93669 Communications equipment, NEC Hardware 43671 Electron tubes Hardware 03672 Printed circuit boards Hardware 53674 Semiconductors and related devices Hardware 453675 Electronic capacitors Hardware 03676 Electronic resistors Hardware 03677 Electronic coils, transformers, and other inductors Hardware 03678 Electronic connectors Hardware 03679 Electronic components, NEC Hardware 43822 Automatic controls for regulating residential and commercial

environments and appliancesMed/lab devices 0

3823 Industrial instruments for measurement, display, and controlof process variables; and related products

Med/lab devices 2

3824 Totalizing fluid meters and counting devices Med/lab devices 03825 Instruments for measuring and testing of electricity and

electrical signalsMed/lab devices 2

3826 Laboratory analytical instruments Med/lab devices 103829 Measuring and controlling devices, NEC Med/lab devices 13841 Surgical, medical, and dental instruments Med/lab devices 63842 Orthopedic, prosthetic and surgical appliances and supplies Med/lab devices 23844 X-ray apparatus and tubes and related irradiation Med/lab devices 03845 Electromedical and electrotherapeutic apparatus Med/lab devices 117371 Computer programming services Software/Internet 417372 Prepackaged software Software/Internet 1047373 Computer integrated systems design Software/Internet 237374 Computer processing and data preparation services Software/Internet 207375 Information retrieval services Software/Internet 297376 Computer facilities management services Software/Internet 07379 Computer related services, NEC Software/Internet 218710 Engineering, architectural, and surveying Analytic services 08711 Engineering services Analytic services 08731 Commercial physical and biological research Analytic services 31

l

IRlcfifw

iatttiat

wt

(p

8732 Commercial economic, sociological, and educationa8733 Noncommercial research organizations

8734 Testing laboratories

n ten cases, we found no official indication of withdrawal via anW form filing. Following SEC regulation, which treats a prospectus

eft unrealized for 270 days as abandoned, we treat those right-ensored “abandoned” IPOs as cases of withdrawal.10 Across allrms we identified, the average time from S-1 filing to success-

ul IPO was 76 days, while the average time from filing to an officialithdrawal was more than double that, 151.5 days.

In our sample, 93% (64) of the firms that filed an S-1 in 1997 real-zed an IPO. This number decreased to 71% in the following year,nd then rose again to 95% in 1999. In 2000, the year when theechnology bubble burst, 76% of companies that filed a prospec-us realized an IPO. Overall, 15% of the sample firms withdrewheir offering after filing an S-1.11 We searched SEC filings (primar-

ly 8-K announcements of unscheduled events), the business pressnd comprehensive databases such as OneSource in order to iden-ify the “final fate” of technology companies that withdrew their

10 Note, however, that eight of the ten cases are firms that had no VC backing orere acquired soon after the withdrawal and thus were excluded from our inferen-

ial analysis.11 While other studies have found an average rate of withdrawal of around 20%Lian and Wang, 2012) those analyses typically include firms that withdraw toursue mergers or acquisitions, which we exclude.

Analytic services 4Analytic services 2Analytic services 2

IPOs. We test our hypotheses by examining the conditions underwhich firms that have filed an S-1 opt to forgo a liquidity event bywithdrawing their offering to remain independent and privatelyowned.

4.3. Independent variables

Our primary independent variables reflect different positions apre-IPO company can occupy in two relevant networks; the struc-ture comprised of syndicated VC investments and the networkmade up of interlocking directorships with publicly traded corpo-rations.

4.3.1. Constructing VC syndicate networksWe constructed four cross-sections of the entire network of ven-

ture capital syndicates. Data were obtained from VentureXpert forall financiers that have made at least one venture capital invest-ment in a U.S. based company. A tie between a venture capitalfirm and a portfolio company is assumed to last until the com-

pany that receives funds goes public, is acquired or receives a newround of venture capital financing. For investment relationshipswith missing end dates, we assume that the tie lasts for five years.We base our measures on complete cross-sections of the network

8 cial Ne

avctI12

4

apdpbcvdiwnosida

loNbfvff2

dnmffi(4ifitwsi

4

msdtsiastafi

J. Owen-Smith et al. / So

s of January 1 of the relevant year. Because withdrawals are onlyery rarely repeated events, our data are structured as a series ofross-sections. In 1997 the two-mode network included 1623 ven-ure capital firms with investments in 7606 portfolio companies.n 1998, 1814 VC firms owned shares in 8604 companies; in 1999,961 VCs had private equity in 9532 corporations. Finally in the year000, 2391 VCs held investments in 10,774 private companies.

.3.2. Constructing the interlock networksConstructing complete interlock networks that include firms

t IPO (and, most importantly, those that never succeed in goingublic) posed a more substantial challenge. There is no existingatabase that comprehensively tracks corporate directors at theoint when a still privately owned firm files for an IPO. Moreover,oard composition often shifts when a firm’s ownership structurehanges. To identify the firms’ connections to public companiesia board interlocks, we collected data on the firms’ officers andirectors from the management and director biographies included

n filed S-1 forms. In total we collected names, ages and positionsithin the firm for 9456 officers and directors of pre-IPO compa-ies. Because many of these companies eventually withdrew theirfferings, relevant director information never became available intandard data sources used for the construction of public companynterlock networks. We thus matched the names of executives andirectors from pre-IPO firms to a larger dataset tracking directorsnd officers for all publicly traded firms.

We constructed four cross-sections of the entire board inter-ock network among U.S.-incorporated public corporations listedn the New York Stock Exchange, American Stock Exchange, andasdaq (Non-national and National systems). Data on board mem-ership were collected from Compact Disclosure’s January releasesor 1997–2000. This set comprises 1045 companies and 46,505 indi-idual director listings for 1997; 7284 firms and 50,328 directorsor 1998; 6855 companies and 55,527 individual director listingsor 1999 and 6525 companies and 49,501 director listings for000.

The network construction process involved identifying whichirectors sat on more than one board, resulting in a two-modeetwork of directors and companies. This was accomplished byatching names and ages of directors, complemented by identi-

ying numbers drawn from multiple sources including SEC insiderlings, Compustat and the Center for Research in Security PricesCRSP), when available. The name-matching process identified548 interlocked companies (plus their directors) in 1997; 4968

n 1998; 5159 in 1999 and 4831 in 2000. We positioned pre-IPOrms in these networks by appeal to the data on officers and direc-ors collected from S-1 forms. Sample firms and their directorsere inserted into the two-mode board interlock network via the

ame name matching procedures, resulting in yearly networks thatnclude both public boards and pre-IPO companies.

.3.3. Measuring the potential for information transfer: pipesTo identify the effects of the “pipes,” or information transfer

echanism, we constructed a modified network centrality mea-ure that we call attention weighted degree centrality. The standardegree centrality measure is defined as the number of connectionshat a company has in a given time. In our context, this mea-ure translates into the number of venture capital firms that havenvestment ties to the company or the number of all directors on

company’s board. We weight this measure by the inverse of the

um of other partners connected to ego’s alters in order to capturehe extent to which a pre-IPO company’s VC investors or directorsre able to focus their attention on the needs and situation of therm. Degree weighted centrality thus captures the exclusivity of a

tworks 41 (2015) 1–17

board’s attention to ego by weighting the number of directors bytheir other connections. We define weighted degree centrality as

dw =k∑

i=1

1di

where di represents the number of connections for the ith firmwith investments in our focal company (or the number of boardson which the ith director sits). Larger values on this measure thusreflect firms that have more partners who themselves are less wellconnected and thus more exclusively focused on ego. For instance, ifa pre-IPO company had six directors on its board, each of whom hadno other directorships at publicly traded corporations, the value ofour pipes measure would be six. If each of those individuals heldthree additional directorships, the value would be 1.5. More suc-cinctly, higher values on this measure indicate that ego has moreexclusive access to the attention of its partners.

4.3.4. Measuring the potential for status signals: prismsWe measure the social status of a pre-IPO firm using eigen-

vector centrality. Where our pipes measure indexes the extentto which an ego must compete with degree two partners for analter’s attention, this measure captures the extent to which anego is made visible by being part of a long chain of connectionslinking prominent alters. In the VC syndicate network we opera-tionalize this measure as the average eigenvector centrality of apre-IPO company’s VC investors in the complete syndicate network.To obtain the eigenvector centrality of VC investors, we created aone-mode investor by investor projection of the syndicate network,where ties between investors indicate sharing an investment in aportfolio company. Similarly, for the interlock network we mea-sure the average eigenvector centrality of our focal firm’s directorsin the director by director projection of the public interlock net-work (where ties between directors indicate co-serving on a firm’sboard). As a result, the prisms measure indexes the amount of sta-tus the firm’s VC investors and board directors bring to the firm. Weexpect increasing status to decrease the likelihood of withdrawalin both networks.

It is important to note the differences between our pipes andprisms measures. While degree and eigenvector centrality are oftenhighly correlated, our emphasis on attention weighted degree cen-trality results in more separable measures. When a pre-IPO firm’ssecond-degree relations (partners of its VC investors or direc-tors) increase, the attention-weighted degree measure decreases,because it is constructed to capture the amount of focus the pre-IPOfirm receives from its first-degree partners (investors and direc-tors). On the other hand, when a pre-IPO firm’s second-degreerelations increase, the eigenvector centrality of the firm is likelyto increase, particularly if those second-degree partners are them-selves prominent. As a result, one would expect these pipes andprisms measures to be negatively correlated. However, the amountof correlation could vary considerably, and would depend on howmuch the number of second-degree partners increases as a functionof the number of first-degree partners. Different networks have dif-ferent norms and rules for attachment. For example, while boardsize can vary highly, even low status firms must have a full com-plement of board members; thus director focus (pipes) and status(prisms) should not be expected to strongly correlate.

4.3.5. Measuring the potential for influence: peersWe measure the potential for social influence by identifying

partners that themselves have prior experience with an IPO with-drawal. Pre-IPO companies whose partners have more collectivewithdrawal experience are more likely to view withdrawal as alegitimate alternative to realizing an IPO. In order to calculate this

cial Ne

mtiffne

scyaowtvhih

4

stfytdtdbtefioqhh

fiitFiasnefi

ipIdrft

bcr

fi

J. Owen-Smith et al. / So

easure we draw on information on IPO withdrawals from the fullime series of electronically available SEC filings (1993–2005). Wedentify a VC firm as having experienced an IPO withdrawal if a port-olio company in which it invested in the prior five years12 filed aorm RW with the SEC. We operationalize this variable as the totalumber of investors in a portfolio company that have withdrawalxperience.

In contrast, a director has relevant experience when he orhe has served on the board of another pre-IPO high technologyompany that withdrew its prospectus during the preceding fiveears. The distribution of directors with withdrawal experiencemong sample firms is highly skewed. Fully 103 of the firms webserve have a director who has experienced a prior technologyithdrawal, while just 18 have two or more such directors. We

hus operationalize our director experience measure as a dummyariable indicating whether the focal firm has any directors whoave experienced an IPO withdrawal. We expect both director and

nvestor experience with IPO withdrawals to increase the likeli-ood a technology company will step away from its IPO.

.4. Control variables

All models include control variables for characteristics of theample firms, the offering and key features of the market con-ext, as well as yearly fixed effects.13 We constructed a variableor firm age (logged) using Loughran and Ritter’s (2004) foundingear data. Founding year is defined as the year of initial incorpora-ion. For spin-offs, founding year is defined as the founding of theivision. Missing data was collected directly from the IPO prospec-us using the same criteria. From the prospectus, we also collectedata on industry and geographical location. We control for industryy constructing four dummy variables for the five general sectorshat comprise our sample. The small analytic services sector whichncompasses contract laboratory, testing, and engineering researchrms serves as our reference category in all reported models. Datan location was used to construct a dummy variable for firms head-uartered in the two states, California and Massachusetts, that areome to the greatest concentration of both venture capital andigh-technology firms (cf. Saxenian, 1994; Powell et al., 2002).

We also control for the financial characteristics of the issuingrm, since such factors may influence both the reservation value of

nsiders and financial backers and signal the firm’s potential valueo investors in the public market (Busaba et al., 2001; Zhao, 2005).irms with greater (or any) revenues tend to be less uncertainnvestment objects and receive higher valuations. The literaturelso proposes a positive association between debt and withdrawalsince organizations that carry significant debt have access to alter-ative sources of capital (Busaba et al., 2001). We control for theseffects by including logged measures of revenue and debt from thenancial information reported in the IPO prospectus.

A second set of variables control for characteristics of the offer-ng. From the SDC New Issues database, we obtained data for thelanned primary use of the proceeds. The data was verified in the

PO prospectus. We also used the firms’ IPO prospectus to collect

ata for firms not included in the SDC New Issues database. Prioresearch has shown that companies that plan to use the proceedsor recapitalization are more likely to withdraw the offering sincehey often have alternative financing (Busaba et al., 2001; Zhao,

12 For S-1s filed in 1997, we are forced to shorten that time period to four yearsecause SEC filing data is left censored at 1993. This difficulty was experienced withalculations for both the syndicate and the interlock networks. Our findings areobust to using a four year window for all experience calculations.13 Because our data structure does not include repeated observations of the samerms across years, fixed firm effects are inappropriate.

tworks 41 (2015) 1–17 9

2005; Dunbar and Foerster, 2008). Thus we recoded the data into abinary indicator for debt payment. Prior research on IPOs has alsopaid extensive attention to the endorsement, or certification, effectof underwriters. We control for this effect by including the Carterand Manaster (1990) underwriter reputation score (updated byCarter et al., 1998; Loughran and Ritter, 2004). The measure is basedon the underwriter’s position in tombstone announcements andit aims to capture the hierarchical structure of investment bank-ing (Podolny, 1993; Jensen, 2003). We code missing values as zeroassuming that underwriters that are not covered by the rankinghave low reputation. Similarly, self-underwritten IPOs are assignedan underwriter reputation of zero since those firms do not benefitfrom any underwriter status effects.

Finally we control for features of the market for technology IPOs.Because competition for the attention of stock market investors isan important factor in IPO success (Gulati and Higgins, 2003) andbecause competition among similar actors increases social compar-ison processes (Burt, 1997; Pollock et al., 2008), we also include avariable that captures the number of firms in the same SIC that are atrisk for IPO or withdrawal. We next calculate a retrospective mea-sure of market uncertainty, which we measure using the numberof S-1s that were withdrawn in the 90 days prior to a company’sown S-1 filing date. Control variables are highly inter-correlatedand their inclusion in these models introduced untenable levelsof multicollinearity. We thus orthogonalize all control measures.Tables 2 and 3 present descriptive statistics and correlations foruntransformed measures.

4.5. Model specification

We present findings from models that use a probit specifica-tion to estimate the likelihood that a pre-IPO firm will withdrawits IPO. We include only those firms that either realized their IPOor withdrew to remain independent and privately owned. Contin-ued independence after withdrawal represents the clearest caseof a pre-IPO company choosing a costly alternative to its statedpreference because in these circumstances a withdrawn offeringleads neither to debut on public equity markets nor to liquiditythrough private merger or acquisition. We run all models withrobust standard errors on the subsample of organizations thatreceived at least one VC investment and did not withdraw theirIPO due to bankruptcy, merger, or acquisition. All continuous inde-pendent variables are mean centered.

5. Findings

The models reported in Table 4 estimate the probability of IPOwithdrawal from 1997–2000.

Firm, offering, market condition controls, and yearly fixedeffects exert stable and largely unsurprising influences on thelikelihood of withdrawal. When yearly variation is accounted for,increasing numbers of withdrawals in the prior quarter have a sig-nificant and positive effect on the likelihood a firm will withdrawits IPO. Likewise, and in line with past findings (Busaba et al., 2001)firms that indicate they will use part or all of their IPO revenuesto pay down accumulated debt have a greater likelihood of with-drawal. Firms with greater revenues are marginally less likely towithdraw their IPOs, a finding that is somewhat surprising as firmswith stronger financial track records typically expect (and receive)higher valuation at IPO.

We suspect that this relatively weak effect may be a function of

our sample. In this time period it was quite common for technol-ogy companies to pursue public offerings with little or no recordof financial success and few revenues. We also find surprising therelatively weak role that underwriter status plays in explaining

10 J. Owen-Smith et al. / Social Networks 41 (2015) 1–17

Table 2Descriptive statistics.

Variables Mean S.D. Min Max

(1) Withdrawal 0.152 0.360 0 1(2) Underwriter reputation (Sqrt) 2.888 0.187 1.761 3.017(3) Competition (Sqrt) 4.869 2.130 1 9.274(4) Age (logged) 1.880 0.624 0 4.533(5) Debt (logged) 2.628 1.119 0.265 7.531(6) Withdrawals, past 90 days (Sqrt) 4.178 1.565 1 9(7) Revenue (logged) 2.161 1.333 0 7.393(8) State cluster 0.563 0.497 0 1(9) Industry: drug and biotech 0.091 0.287 0 1(10) Industry: hardware 0.223 0.417 0 1(11) Industry: medical/lab devices 0.075 0.264 0 1(12) Industry: software and Internet 0.525 0.500 0 1(13) Year 1998 0.130 0.337 0 1(14) Year 1999 0.360 0.480 0 1(15) Year 2000 0.358 0.480 0 1(16) Proceeds to pay debt 0.108 0.311 0 1(17) Average VC eigenvector centrality (VC Prisms) 0.000 0.027 −0.048 0.111(18) Average director eigenvector centrality (Int Prisms) 0.000 0.349 −0.068 5.280(19) Attention weighted degree centrality, VC Net (VC Pipes) 0.000 0.650 −0.499 5.369(20) Attention weighted degree centrality, Interlock Net (Int Pipes) 0.000 1.642 −3.697 8.470(21) VC withdrawal experience (VC Peers) 1.947 1.904 0 10(22) Director withdrawal experience (Int Peers) 0.267 0.443 0 1(23) VC Pipes × Prisms −0.004 0.015 −0.092 0.055

tawrncits

sssndwwatwnpds

eashisIgVsd

i

(24) Int Pipes × Prisms

(25) Int Peers × Pipes(26) VC Peers × Prisms

he likelihood of withdrawal. The underwriter status variable has marginally significant and positive effect on the likelihood ofithdrawal, but only in models 4 and 7, which include the full

ange of measures and contingencies for both interlock and VCetworks. There are few robust industry effects. Medical Deviceompanies are marginally more likely to withdraw than are firmsn the Analytic Services sector, which is our omitted category. Withhe exception of underwriter status, control effects are qualitativelytable across specifications.

Consider Model 7, which estimates effects for all hypothe-ized direct effects and interactions. An initial look at this modeluggests partial support for several of our hypotheses. Generallypeaking, our arguments are more strongly supported in the VCetwork than in the interlock network. Hypothesis 2, which pre-icted that stronger status signals sent through network ‘prisms’ould decrease the likelihood of withdrawal, holds in the VC net-ork but not in the interlock network. Hypothesis 1, which made

similar prediction that increased access to valuable informa-ion and resources through more exclusive network ‘pipelines’ould decrease the likelihood of withdrawal, also plays out for VCetworks but not for interlock networks. This model offers no sup-ort for our third hypothesis. Connections to either investors orirectors who have experienced a prior IPO withdrawal have noignificant effect on the likelihood of withdrawal.

Already it is clear that different, potentially relevant, networksxert different effects. Indeed, we find it very surprising that therere so few effects to be found for the director network. One pos-ible explanation for the lack of effects is that outside directorsave less input into strategic decisions about IPOs than do VC

nvestors. Outside directors, for instance, may play a primarily advi-ory role and have relatively little financial stake in a particularPO, while venture capitalists are very hands on and often have areat deal invested. Another, complementary explanation is that

C and director interests are largely aligned as lead VCs exertome degree of control over board membership and often occupyirectorships.14 Of course some suggest that directors play a purely

14 Indeed, our interlock pipelines measure would have had a positive and signif-cant effect on a two tailed test, suggesting that as a firms’ directors focus their

−0.035 0.420 −7.201 1.591−0.031 0.748 −2.497 4.970

0.025 0.061 −0.138 0.308

symbolic role in modern public corporations (Davis and Robbins,2005), which might also explain the surprising lack of “director”effects on this very important decision. The differences betweenthese two networks become even more apparent when we considerhypothesized contingencies.

Hypothesis 4 predicts a positive interaction between pipe andprism mechanisms in the interlock network because the obliga-tions directorships place on individuals suggest that as the averagestatus of a company’s directors increases, their ability to transfertimely and appropriate information and resources will decrease. Inthe terms we use above, as increasing connections tax the abilityof individual directions to attend to ego, the competitive bene-fits of information flows through partners diminishes. Here weexpect any positive effects of status signaling to be offset as dif-fuse attention of directors hampers the timely flow of valuableinformation.

While care should be taken in interpreting a significant inter-action where no predicted direct effects are found, positive andsignificant effects in Models 4 and 7 are very suggestive. Moreattention should be paid to the possibility that the kinds of obliga-tions ties place on alters and limitations on alters’ ability to attendeffectively to all their partners can create conditions where positivenetwork effects dampen each other.

In similar vein, Hypothesis 5 predicts a negative interactionbetween pipe and prism mechanisms in VC network because thestandard business practices of VCs make it possible for very wellconnected investors to also attend very closely to individual port-folio companies. Unlike the director network, increasing visibilityin the VC network should increase access to valuable informationand resources and the joint effect of pipes and prisms will be tomake the likelihood of withdrawal even more remote. In the termswe mobilize earlier, the quality of attention VC made possible bythe structure of VC partners results in a situation where advantages

available to ego through pipe and prism effects are cumulative.Here we expect the competitive advantages associated with sta-tus signals to amplify those that accrue to information flows as

attention more exclusively on ego by maintaining fewer simultaneous directorshipsat other public companies, the likelihood of withdrawal increases.

J. O

wen-Sm

ith et

al. /

Social N

etworks

41 (2015)

1–17

11

Table 3Bivariate correlations.

Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26)

(1) Withdrawal 1.000(2) Underwriter

reputation (Sqrt)0.051 1.000

(3) Competition(Sqrt)

0.056 0.164 1.000

(4) Age (logged) 0.000 −0.136 −0.005 1.000(5) Debt (logged) −0.012 0.166 0.012 0.083 1.000(6) Withdrawals, past

90 days (Sqrt)0.125 0.025 0.103 −0.013 0.133 1.000

(7) Revenue (logged) −0.072 0.032 0.017 0.314 0.596 −0.001 1.000(8) State cluster −0.085 0.028 −0.039 −0.038 −0.060 −0.046 −0.135 1.000(9) Industry: drug

and biotech0.016 −0.178 −0.329 −0.034 −0.058 0.046 −0.135 −0.017 1.000

(10) Industry:hardware

−0.079 0.039 −0.263 0.091 0.203 0.007 0.251 0.076 −0.169 1.000

(11) Industry:medical/labdevices

0.066 −0.081 −0.354 0.058 −0.136 0.019 −0.164 0.116 −0.090 −0.153 1.000

(12) Industry:software andInternet

0.034 0.129 0.779 −0.066 −0.041 −0.063 0.037 −0.133 −0.332 −0.564 −0.300 1.000

(13) Year 1998 0.146 −0.026 −0.223 −0.013 0.031 −0.040 0.135 −0.095 0.015 −0.065 −0.011 0.092 1.000(14) Year 1999 −0.215 0.063 0.242 −0.061 −0.040 −0.352 −0.033 0.067 −0.140 −0.059 −0.161 0.289 −0.290 1.000(15) Year 2000 0.184 0.081 0.162 0.034 0.106 0.494 −0.087 −0.002 0.118 0.032 0.067 −0.222 −0.289 −0.559 1.000(16) Proceeds to pay

debt0.090 −0.129 −0.036 0.064 0.104 −0.070 0.152 −0.080 −0.011 0.035 −0.018 0.018 0.119 −0.009 −0.171 1.000

(17) Average VCeigenvectorcentrality (VCPrisms)

−0.070 0.186 −0.048 −0.033 −0.046 0.000 −0.027 0.222 −0.004 0.046 0.031 −0.051 0.005 −0.008 −0.006 −0.080 1.000

(18) Average directoreigenvectorcentrality (IntPrisms)

0.019 −0.026 −0.047 −0.060 0.103 0.000 0.091 −0.046 −0.023 0.001 −0.035 0.026 0.344 −0.134 −0.046 0.048 0.017 1.000

(19) Attentionweighted degreecentrality, VC net(VC Pipes)

−0.051 0.024 −0.070 0.036 −0.059 0.022 −0.168 0.139 0.085 0.069 −0.025 −0.105 −0.043 −0.029 −0.013 −0.066 −0.213 −0.006 1.000

(20) Attentionweighted degreecentrality,interlock net (IntPipes)

0.142 0.016 −0.026 0.096 0.056 −0.042 0.001 −0.206 0.055 −0.027 −0.036 −0.020 −0.008 −0.053 0.044 0.017 −0.177 −0.061 0.086 1.000

(21) VC withdrawalexperience (VCPeers)

−0.059 0.137 −0.024 −0.003 −0.065 0.072 −0.121 0.247 0.033 0.090 −0.023 −0.112 −0.041 0.040 0.067 −0.095 0.474 0.037 0.197 −0.109 1.000

(22) Directorwithdrawalexperience (IntPeers)

−0.034 0.014 0.011 0.070 −0.003 0.132 −0.102 0.099 0.070 0.012 −0.002 −0.086 −0.130 0.005 0.153 −0.050 0.058 0.019 0.115 −0.043 0.318 1.000

(23) VC Pipes × Prisms −0.004 −0.002 0.075 0.009 0.080 0.109 0.111 −0.073 −0.103 0.056 −0.062 0.040 −0.007 −0.070 0.068 0.026 −0.270 −0.025 −0.261 −0.068 0.041 0.064 1.000(24) Int Pipes × Prisms 0.040 0.047 0.045 0.004 −0.045 −0.005 −0.057 0.067 0.020 −0.004 0.018 −0.018 −0.213 0.077 0.051 0.014 −0.022 −0.775 0.000 −0.092 −0.053 −0.015 0.040 1.000(25) Int Peers × Pipes 0.014 0.005 −0.036 0.102 0.026 −0.010 0.056 −0.115 0.022 −0.024 0.035 −0.020 0.012 0.037 −0.032 −0.070 −0.099 −0.046 0.071 0.456 −0.088 −0.069 −0.056 −0.021 1.000(26) VC Peers × Prisms 0.003 0.128 −0.073 0.013 −0.046 −0.001 −0.047 0.216 −0.011 0.072 0.040 −0.090 0.008 0.012 −0.027 −0.085 0.714 0.081 −0.157 −0.118 0.605 0.082 −0.139 −0.088 −0.091 1.000

12

J. O

wen-Sm

ith et

al. /

Social N

etworks

41 (2015)

1–17

Table 4Probit models on the likelihood of withdrawal.

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

Control variablesUnderwriter reputation (Sqrt, orthogonalized) (SE) 0.091 (0.082) 0.112 (0.083) 0.117 (0.085) 0.141+ (0.083) 0.119 (0.085) 0.133 (0.086) 0.147+ (0.083)Competition (orthgonalized) 0.087 (0.074) 0.067 (0.073) 0.074 (0.074) 0.083 (0.077) 0.075 (0.075) 0.087 (0.076) 0.091 (0.078)Age (logged, orthogonalized) 0.048 (0.085) 0.038 (0.085) 0.033 (0.085) 0.017 (0.086) 0.034 (0.085) 0.02 (0.086) 0.011 (0.087)Debt (logged, orthogonalized) −0.044 (0.083) −0.051 (0.084) −0.073 (0.087) −0.07 (0.089) −0.075 (0.087) −0.066 (0.088) −0.063 (0.090)Withdrawals, past 90 days (Sqrt, orthogonalized) 0.240** (0.074) 0.251*** (0.073) 0.279*** (0.075) 0.300*** (0.075) 0.290*** (0.074) 0.303*** (0.074) 0.309*** (0.074)Revenue (logged, orthogonalized) −0.129 (0.083) −0.127 (0.084) −0.146+ (0.088) −0.164+ (0.089) −0.152+ (0.089) −0.135 (0.089) −0.147+ (0.089)State cluster (orthogonalized) −0.169* (0.078) −0.136+ (0.079) −0.078 (0.087) −0.068 (0.087) −0.073 (0.087) −0.078 (0.090) −0.076 (0.090)Drugs/biotech (orthogonalized) 0.041 (0.076) 0.046 (0.076) 0.057 (0.080) 0.051 (0.081) 0.06 (0.080) 0.064 (0.083) 0.056 (0.082)Hardware (orthogonalized) −0.106 (0.080) −0.104 (0.079) −0.093 (0.081) −0.086 (0.082) −0.092 (0.081) −0.089 (0.082) −0.086 (0.082)Med/lab devices (orthogonalized) 0.128+ (0.069) 0.127+ (0.069) 0.133+ (0.070) 0.129+ (0.071) 0.132+ (0.070) 0.131+ (0.070) 0.128+ (0.071)Software/Internet (orthogonalized) −0.014 (0.076) −0.005 (0.076) 0.011 (0.077) 0.001 (0.079) 0.003 (0.077) 0.003 (0.077) 0.005 (0.078)Year 1998 (orthogonalized) 0.327*** (0.080) 0.353*** (0.083) 0.355*** (0.083) 0.367*** (0.083) 0.354*** (0.082) 0.366*** (0.085) 0.367*** (0.086)Year 1999 (orthogonalized) −0.270** (0.091) −0.284** (0.089) −0.281** (0.090) −0.297** (0.092) −0.276** (0.090) −0.285** (0.092) −0.306** (0.094)Year 2000 (orthogonalized) 0.111 (0.081) 0.115 (0.081) 0.109 (0.082) 0.102 (0.084) 0.112 (0.083) 0.127 (0.084) 0.114 (0.084)Proceeds to pay debt (orthogonalized) 0.192** (0.070) 0.194** (0.071) 0.190** (0.070) 0.186* (0.073) 0.191** (0.070) 0.193** (0.072) 0.185* (0.075)

Independent variablesAverage VC eigenvector centrality (VC Prisms) −5.058+ (3.130) −5.708* (3.310) −10.418** (3.808) −4.975 (3.949) −10.619* (4.571) −15.765** (5.309)Average director eigenvector centrality (Int Prisms) −0.18 (0.199) −0.098 (0.169) 0.132 (0.296) −0.086 (0.169) −0.174 (0.166) 0.092 (0.289)Attention weighted degree, VC net (VC Pipes) −0.270* (0.163) −0.563** (0.200) −0.249+ (0.173) −0.183 (0.172) −0.513* (0.246)Attention weighted degree, Interlock net (Int Pipes) 0.133 (0.047) 0.144 (0.049) 0.131 (0.048) 0.140 (0.053) 0.153 (0.054)VC Pipes × prisms −16.777* (7.831) −15.259* (9.016)Int Pipes × Prisms 0.651* (0.371) 0.682* (0.375)VC withdrawal experience (VC Peers) −0.016 (0.056) −0.08 (0.063) −0.007 (0.072)Director withdrawal experience (Int Peers) −0.093 (0.186) −0.055 (0.189) −0.091 (0.191)Int Peers × Pipes −0.065 (0.131) −0.055 (0.135)VC Peers × Prisms 5.315** (2.095) 4.159* (2.091)

Constant −1.212*** (0.089) −1.229*** (0.089) −1.268*** (0.094) −1.352*** (0.103) −1.215*** (0.147) −1.254*** (0.146) −1.421*** (0.178)

Observations 453 453 453 453 453 453 453Adjusted R-squared 0.149 0.157 0.181 0.2 0.182 0.197 0.211

+ p < 0.1.* p < 0.05.

** p < 0.01.*** p < 0.001 (one-tailed tests for independent variables, two-tailed tests for controls).

cial Ne

pct

wifai6odostemwsid

tnnfiaoIeFmdCdco

imotaawIbNiavcp

ssripiofdpbw

the effects exerted by director and VC connections. Increasing sta-tus in the interlock network from the 1st to 5th quintile changesthe likelihood of withdrawal by only a small amount when the

J. Owen-Smith et al. / So

rominent VC partners send stronger signals while being able tohannel more and better information to ego. Models 4 and 7 supporthis contention.

Finally, consider our hypotheses about the conditions underhich partners’ experiences of withdrawal will become salient

n a pre-IPO firm’s decision making. The lack of support we findor Hypothesis 3 suggests that having directors or investors whore experienced with IPO withdrawals by itself is not sufficient tonfluence a technology firm’s decision to withdraw. In Hypotheses

and 7 we expand our arguments about the distinctive featuresf director interlock and VC syndication networks to make pre-ictions about the conditions under which each of these typesf partners’ experience with other IPO withdrawals will becomealient enough to influence ego’s decisions. We rely on the ideahat increasing dependence on a partner makes their views andxperience more influential to ego to predict that having theore exclusive attention of directors will increase interlock net-ork peer effects on the likelihood of withdrawal (H6). By the

ame token we predict that having higher status investors willncrease the VC network peer effects on the likelihood of with-rawal (H7).

Hypothesis 6 is not supported in our models. In keeping with ourhoughts about the surprising lack of support for other interlocketwork hypotheses, we believe that this is related to the advisoryature of outside directors’ role in the strategic direction of therm. Hypothesis 6 may not be supported because more exclusivettention to a focal company may not be the most important sourcef a director’s power and thus of the salience of their experiences.nstead, we suggest that other processes may determine the influ-nce of directors in privately held firms such as those in this sample.or example, directors who own significant shares of stock may beore clearly influential in decision-making. Likewise, independent

irectors may be more salient in decision making on boards whereEOs do not also hold the position of chairman, by virtue of theirectors’ increased authority and independence. Because our con-eptual focus is on blends of network mechanisms, we test neitherf these possibilities.

Hypothesis 7 does find support in Model 7. As the status of VCnvestors increases, increasing experience with prior withdrawals

akes technology companies more likely to withdraw an IPO. Inther words, prism-like status effects make the experience of ven-ure capitalists more influential. The blend of prism and peer effectspparent in the VC network suggests that higher status financiersre better able to turn experience with IPO withdrawal towardithdrawal, continued independence and the possibility of a later

PO. The negative reputational consequences of withdrawal mighte buffered by the work of experienced and high status investors.evertheless, in the VC network status increases influence, result-

ng in a situation where having partners who are experienced withlternatives and whose other connections more clearly signal one’salue to potential investors actually makes it less likely that aompany will realize an IPO despite having expressed a strongreference to do so.

In Table 4, we find mixed support for our direct effect hypothe-es and three of our four hypotheses about network blends areupported. VC status increases the salience of financiers’ expe-ience with alternatives to IPO. Increased status and access tonformation via the more exclusive attention of partners are com-lementary in the VC network but work at cross purposes in the

nterlock network. The picture here is complicated but supportsur overarching claim that the practical effects of networks resultrom blends of mechanisms that moderate one another differently

epending on the character of the partners and ties that make up aarticular structure. Different networks exert disparate effects onehavior and outcomes because they mix mechanisms in distinctays.

tworks 41 (2015) 1–17 13

6. Discussion

We present graphs of predicted probabilities of withdrawal fromModel 7 as a means to ground our discussion of the larger implica-tion of theorizing mechanism blends. We begin by considering theclearest instance where two distinct networks manifest differentblends of pipe and prism effects. Fig. 1 presents the predicted prob-abilities of withdrawal for each quintile of our attention weighteddegree (pipes) and eigenvector centrality (prisms) mechanisms forthe VC syndication network. Predicted probabilities of withdrawalfor each quintile were calculated with control variables and othercontinuous independent variables held at their means15 and dis-crete independent variables held at their medians. Fig. 2 does thesame for the interlock network, with predicted probabilities ofwithdrawal for each quintile of the interlock attention weighteddegree and interlock eigenvector centrality variables.

In these figures, higher values suggest a greater likelihood thata young technology company will withdraw its IPO. Fig. 1 tellsan interesting story. Companies with the lowest status VCs (thehistogram cluster on the left) see an increasing likelihood of with-drawal as those VCs have fewer other active investments. The tallestbar in this figure represents a 23% probability of withdrawal, whichis for firms whose relatively low average status financiers attendmost exclusively to ego. In contrast, the bars in the rightmost his-togram cluster, which represent companies with the backing ofthe highest status VCs, show the opposite effect. As the attentionof higher status VCs gets more exclusive, it becomes increasinglylikely that a technology firm will realize its IPO. The shortest blackbar in Fig. 1 reflects this scenario, with a less than 1% likelihood ofwithdrawal. Put more simply, the pre-IPO companies that are mostlikely to withdraw their IPOs are those with low status investorsthat buy stakes in relatively few other companies. In contrast,those least likely to withdraw are high status VCs that attend rela-tively exclusively to their portfolio companies. Here pipe and prismeffects amplify each other.

For companies with low VC status (leftmost histogram cluster),the difference between having investors with exclusive attentionversus those whose efforts are spread across many investments isabout a 4.5% increase in the likelihood of withdrawal (from 18.8% to23.3%). When VC backers have high status (the rightmost histogramcluster) however, the difference between VCs in the 1st quintile forexclusive attention and in the 5th quintile is a 5% lower likelihoodof withdrawal. The clear message of Fig. 1 is that in the syndicatenetwork, increasing VC status (prism effects) makes investor atten-tion (pipes effects) beneficial for a portfolio company’s ability toachieve its stated goals. In contrast, decreasing status makes thesame level of attention detrimental, increasing the probability thata technology company will forgo its IPO after filing a prospectus.

The story is different in the interlock network,16 where the thinreed of individual directors’ capacity for engagement with the focalfirms leads to tradeoffs between status and attention. Fig. 2 reportsthe predicted probability of withdrawal for pre-IPO companies foreach quintile of our interlock status (prisms) and weighted degree(pipes) measures. All control variables and other independent vari-ables are set to their means and medians.

In Fig. 2, a significant interaction between pipes and prismsmeasures in the interlock measures reinforces differences between

15 Note that control variables are orthogonalized. The results of orthogonalizationare continuous mean-centered variables, even for controls that were count variables.

16 Though we note that an unhypothesized positive direct effect of VC pipes dom-inates this figure, so caution should be taken in its interpretation.

14 J. Owen-Smith et al. / Social Networks 41 (2015) 1–17

gree c

adtotitipsndtsrtis

i

F

Fig. 1. Predicted probability of withdrawal by quintile of attention weighted de

ttention of directors is constant. What is suggestive here is theifference between this image and Fig. 1. Here, having more direc-ors who serve exclusively on ego’s board increases the likelihoodf withdrawal at all status levels. However, consider the lowestwo quintiles of director exclusivity. As average director statusncreases, the rate of withdrawal decreases by a small amount. Onhe other hand, at the highest level of director exclusivity, increasesn status instead result in higher withdrawal rates. Where pipes andrisms work conjointly in the VC syndicate network, our evidenceuggests that they work at cross purposes in the director interlocketwork, further supporting the spirit of our effort to conceptualizeifferent networks in terms of the obligations ties exert on par-icipants of different capacities to meet them. More generally, inome networks, the character of participants and the obligations ofelationships lead increases in status to impose tradeoffs on accesso information. In other networks, different configurations allow

ncreasing status to amplify the competitive benefits gained viaearch through networks.

Our results also suggest that the prior relevant experiences ofnvestors have a greater effect on portfolio companies when their

ig. 2. Predicted probability of withdrawal by quintile of attention weighted degree cent

entrality (VC network) (pipes) and average VC eigenvector centrality (prisms).

increased status makes those experiences more salient. ConsiderFig. 3, which reports the predicted probability of withdrawal acrossquintiles of VC status and experience with IPO withdrawal. In bothfigures, the lowest quintile of peers represents companies whoseinvestors have no experience with withdrawal. The second quintilerepresents organizations with one experienced investor, the thirdquintile represents those with two experienced investors, and thefourth quintile represents those with three experienced investors.Finally, the fifth quintile contains firms with four or more with-drawal experienced VC backers.

When investors have no withdrawal experience, increasing sta-tus decreases the likelihood of withdrawal. Portfolio companieswhose withdrawal-inexperienced VCs occupy the lowest statusquintile (the leftmost bar in the leftmost histogram cluster of Fig. 3)have approximately a 21% probability of stepping off the IPO mar-ket. In contrast, those with inexperienced VCs in the high status

quintile (the leftmost bar in the rightmost histogram cluster) haveabout a 2% chance of withdrawing their IPO. When the possibilityof influence is low, then, increasing status makes withdrawal lesslikely.

rality (interlock net) (pipes) and average director eigenvector centrality (prisms).

J. Owen-Smith et al. / Social Networks 41 (2015) 1–17 15

awal

7

ntiasoa

eifintsclfUracntsoea

ctmm

wirn

Fig. 3. Predicted probability of withdrawal by quintile of VC withdr

. Conclusions and implications

These findings imply stronger support for the idea that blends ofetwork mechanisms shape organizational decision-making thanhey do for the direct effect hypotheses about how similar pos-tions in different networks make corporations more competitivend thus better able to achieve their goals. The strongest conclu-ion to be drawn from this work is that network effects on the sameutcome can differ dramatically depending on the mix of partnersnd activities that make up a structure.

It is fairly clear that in the VC syndicate network, prismffects are the dominant note in the blend of mechanisms. Here,ncreasing status alters the character of both information trans-er and influence. Growing status increases beneficial resource andnformation flows from partners to ego, because the typical orga-izational arrangements of VC partnerships make it possible forheir increased network reach and experience to be turned to theervice of particular portfolio companies. These claims are compli-ated by the dual market structure of the field and the associatedikelihood that venture capitalists of different statuses may activelyavor M&A markets as a route to liquidity for their investments.nder such conditions, VC experience with withdrawal may not

eflect the possibility of influence away from ego’s preferred courses much as it indexes financiers’ ability to accomplish the goal ofashing out, which may well be shared by insiders at a pre-IPO tech-ology company. In future research, more attention should be paido differentiation in the goals and types of expertise of partnersuch as VC backers. We note, however, that this need is reflectivef one of the important ideas we propose; that network effects forgo are often as much or more a function of alters’ needs and focis they are of ego’s preferences and strategies.

If prisms are the clear top note of the mechanism blend thatharacterizes the VC syndicate network, the mix of processes inhe interlock network is muddier. The attention of directors clearly

atters, but our findings suggest tradeoffs across status and infor-ation mechanisms in this context that are not apparent for VCs.Pipes and prisms work and blend differently in the two networks

e examine and the difference appears to stem from the burdensncreasing network reach puts on the ability of partners to conveyelevant and timely information to ego. We also found that part-ers’ experiences with withdrawals alone have little effect on the

experience (peers) and average VC eigenvector centrality (prisms).

actions of ego. The influence mechanism, however is not moderatedby attention (the pipes mechanism) as we expected for directors.It may well be that organizational and interpersonal rather thannetwork factors make directors’ experiences more salient to thecorporations on whose boards they serve. Both formal authorityand the diversity and dynamics of the board room may be moreimportant than status or attention in explaining when influenceflows through interlocks. Future research should strive to moreclosely integrate these three levels of analysis.

Our paper highlights the potential power of a theory of networkmultiplicity that emphasizes the different ways in which distinctempirical structures blend and combine characteristic social mech-anisms. Identifying key mechanisms and specifying how they willmix in differently configured networks is a first step toward nailingdown the sources of contingent network effects and the system-atic differences that underpin the complicated implications ofsimultaneous positions held in multiple structures. Focusing onmechanism blends driven by differences in the capacities of part-ners and types of activities that make up particular structuresalso has implications for how we understand, theorize, and studynetworks in a variety of settings.

First and most important, our findings challenge the idea thatnetworks are resources that can be used in fairly straightforwardways by egos anxious to attain particular goals. As we noted earlier,the time period immediately before IPO is one in which orga-nizations like those we study have the most control over theirnetwork, yet our findings suggest that there are circumstanceswhere salutary positions allow partners to exert influence thatmakes accomplishing one’s goals less likely. This is apparent in thetradeoffs we find between pipes and prisms in the interlock net-work and in the complicated relationship between status (prism)and influence (peers) we find in the VC syndicate network.

In the VC network, it is clearly the case that gaining a connectionto high status partners increases visibility and information accessbut also comes at the cost of autonomy as the very benefits of con-nection to high status alters increase ego’s dependence and thusmake partners’ concerns more salient and telling. A key implica-

tion of our findings is that the effects networks exert can havelittle to do with the uses to which ego tries to put them. Partnersmatter immensely and our theories have done too little to exam-ine the ways in which characteristics of alters shift the likelihood

1 cial Ne

aiftip

witofbot

bacfnofrntotasafoqa

tttembtos

tistncftmtgto

R

A

BB

6 J. Owen-Smith et al. / So

nd effects of ties ego seeks to forge. Our work suggests that stud-es concerned with mapping and explaining the dynamics of tieormation and network evolution should find ways to attend tohe goals and characteristics of both parties to a potential tie; thempact of particular connections depends on the activities of botharties.

In short, this work has three key implications for studies of net-ork dynamics and effects. First, emphasizing mechanism blends

mplies a need to map and characterize the variety of processeshrough which observed connections translate into behaviors andutcomes. We emphasize three – information and resource trans-er (pipes), status signaling (prisms) and social influence (peers) –ut other mechanisms may be at work in the structures that areur focus and processes we do not consider here may be the key tohe functioning of other types of relevant networks.

Second, our findings further support a move toward a moreehavioral and contingent form of network theorizing. Rather thanttempting to identify how networks influence action and out-omes across settings, we believe our findings imply the need toocus on specifying the conditions under which particular mecha-isms and mixes dominate in specific structures and contexts. Inther words, both our theories and our research should move awayrom more structural approaches and toward deeper, more natu-alistic and empirically grounded studies of networks in context. Asoted in the introduction to this paper, there is a tension betweenhis need for contextualization and the need for generalizable the-ry. Finding an appropriate balance requires explicitly consideringhe dynamics of the empirical context during theory development,s was done in this paper. All networks are not alike, and even theame structures may exert different effects as changing conditionslter the mix of mechanisms that make relationships efficaciousor those that participate in them. In addition to changing the focusf our theorizing, this idea suggests the need for more and deeperualitative studies of how people and organizations understand,dapt, inhabit, and use their networks.

Third, our emphasis on social influence and the possibility ofradeoffs among pipe and prism mechanisms underscores the ideahat networks are equivocal strategic resources. Whether desiredies form or not is as much a function of the needs of alters as it is ofgo, and the effects different network positions exert can depend onatters far outside of ego’s control. It is thus difficult if not impossi-

le to unequivocally read an organization or individual’s goals fromheir positions in particular networks. Even when actors manage toccupy what they take to be salutary positions, the link betweentructural location and desired outcomes can be an attenuated one.

Taken together, these implications suggest that approaches thatake contingent and blended network mechanisms as their start-ng point have great potential to illuminate the ways in whichavvy actors use (or are used by) their multiple networks in par-icular settings. Mechanism blends can help explain why differentetworks exert different effects, why the effects of some networkshange as their contexts shift, and why even the most success-ul strategic actors can sometimes find themselves hamstrung byheir connections. In short, we believe that focusing on mixes of

echanisms is a necessary step toward a richer and higher fidelityheory of networks. Such a theory is necessary if we are to inte-rate increasing concerns with network variability and dynamicshat have come to characterize the field with an analytic emphasisn individual and collective action and outcomes.

eferences

cimovic, P., Lyn, E.O., 2000. The anatomy of a failed IPO: the case of Wired Ventures,Inc. Financ. Pract. Educ. Fall/Winter, 14–23.

lau, Peter M., 1964. Exchange and Power in Social Life. Wiley, New York, NY.oeh, K.K., Southam, C., 2011. Impact of initial public offering coalition on deal

completion. Venture Cap.: Int. J. Entrep. Financ. 13, 313–336.

tworks 41 (2015) 1–17

Burt, R.S., 1987. Social contagion and innovation: cohesion versus structural equiv-alence. Am. J. Sociol. 92, 1287–1335.

Burt, R.S., 1992. Structural Holes: The Social Structure of Competition. Harvard Uni-versity Press, Cambridge, MA.

Burt, R.S., 1997. The contingent value of social capital. Admin. Sci. Q. 42, 339–365.Burt, R.S., 2005. Brokerage and Closure: An Introduction to Social Capital. Oxford

University Press, New York, NY.Busaba, W.Y., 2006. Bookbuilding, the option to withdraw, and the timing of IPOs. J.

Corp. Financ. 12, 159–186.Busaba, W.Y., Benveniste, L.M., Guo, R.-J., 2001. The option to withdraw IPOs during

the premarket: empirical analysis. J. Financ. Econ. 60, 73–102.Carter, R.B., Dark, F.H., Singh, A.K., 1998. Underwriter reputation, initial returns, and

the long-run performance of IPO stocks. J. Financ. 53, 285–311.Carter, R.B., Manaster, S., 1990. Initial public offerings and underwriter reputation.

J. Financ. 45, 1045–1067.Coleman, J.S., Katz, E., Menzel, H., 1966. Medical Innovation: A Diffusion Study.

Bobbs-Merrill Co., Indianapolis, IN.Davis, G.F., Robbins, G.E., 2005. Nothing but net? Networks and status in corpo-

rate governance. In: Knorr-Cetina, K., Preda, A. (Eds.), The Sociology of FinancialMarkets. Oxford University Press, Oxford, pp. 290–311.

Dunbar, C.G., Foerster, S.R., 2008. Second time lucky? Withdrawn IPOs that returnto the market. J. Financ. Econ. 87, 610–635.

Fernandez, R.M., Castilla, E.J., Moore, P., 2000. Social capital at work: networks andemployment at a phone center. Am. J. Sociol. 105, 1288–1356.

French, J.R.P., Raven, B., 1959. The bases of social power. In: Cartwright, D. (Ed.),Studies in Social Power. Institute for Social Research, Ann Arbor, MI, pp. 150–167.

Friedkin, N.E., 1993. Structural bases of interpersonal influence in groups: a longi-tudinal case study. Am. Sociol. Rev. 58, 861–872.

Friedkin, N.E., 2003. Social influence network theory: toward a science of strategicmodification of interpersonal influence systems. In: Breiger, K.C., Pattison, P.(Eds.), Dynamic Social Network Modeling and Analysis. National Academy ofSciences/National Research Council Committee on Human Factors, Washington,DC, pp. 89–100.

Friedkin, N.E., Johnsen, E.C., 1990. Social influence and opinions. J. Math. Sociol. 15,193–205.

Galaskiewicz, J., 1985. Professional networks and the institutionalization of a singlemind set. Am. Sociol. Rev. 50, 639–658.

Gompers, P.A., Lerner, J., 1999. The Venture Capital Cycle. MIT Press, Cambridge, MA.Granovetter, M.S., 1973. The strength of weak ties. Am. J. Sociol. 78,

1360–1380.Gulati, R., Higgins, M.C., 2003. Which ties matter when? The contingent effects

of interorganizational partnerships on IPO success. Strat. Manage. J. 24,127–144.

Haunschild, P.R., Beckman, C.M., 1998. When do interlocks matter? Alternate sourcesof information and interlock influence. Admin. Sci. Q. 43, 815–844.

Higgins, M.C., Gulati, R., 2006. Stacking the deck: the effects of top managementbackgrounds on investor decisions. Strat. Manage. J. 27, 1–25.

Hochberg, Y.V., Ljungqvist, A., Lu, Y., 2007. Whom you know matters: venture capitalnetworks and investment performance. J. Financ. 62, 251–301.

Hsu, D.H., 2004. What do entrepreneurs pay for venture capital affiliation? J. Financ.59, 1805–1844.

Jensen, M., 2003. The role of network resources in market entry: commercial banks’entry into investment banking, 1991–1997. Admin. Sci. Q. 48, 466–497.

Kogut, B., Urso, P., Walker, G., 2007. Emergent properties of a new financial market:American venture capital syndication, 1960–2005. Manage. Sci. 53, 1181–1198.

Lave, J., Wenger, E., 1996. Practice, person, socal world. In: Daniels, H. (Ed.), AnIntroduction to Vygotsky. Routledge, London, pp. 143–150.

Lerner, J., 1994. Venture capitalists and the decision to go public. J. Financ. Econ. 35,293–316.

Lerner, J., 2012. Venture capitalists and the oversight of private firms. J. Financ. 50,301–318.

Lian, Q., Wang, Q., 2009. Does the market incorporate previous IPO withdrawalswhen pricing second-time IPOs? Financ. Manage. 38, 357–380.

Lian, Q., Wang, Q., 2012. Acquisition valuations of withdrawn IPOs: when IPO plansturn into mergers. J. Bank. Financ. 36, 1426–1436.

Lin, N., 2001. Social Capital: A Theory of Structure and Action. Cambridge UniversityPress, London and New York, NY.

Loughran, T., Ritter, J., 2004. Why has IPO underpricing changed over time? Financ.Manage. 33, 5–37.

March, J.G., Simon, H.A., 1958. Organizations. John Wiley and Sons, New York.Marsden, P.V., Friedkin, N.E., 1993. Network studies of social influence. Sociol. Meth-

ods Res. 22, 127–151.McDonald, M.L., Westphal, J.D., Graebner, M.E., 2008. What do they know? The

effects of outside director acquisition experience on firm acquisition perfor-mance. Strat. Manage. J. 29, 1155–1177.

Mizruchi, M., 1992. The Structure of Corporate Political Action: Interfirm Relationsand Their Consequences. Harvard University Press, Cambridge, MA.

Mizruchi, M.S., 1996. What do interlocks do? An analysis, critique, and assessmentof research on interlocking directories. Ann. Rev. Sociol. 22, 271–298.

Palmer, D., Barber, B., Zhou, X., Soysal, Y., 1995. The friendly and predatory acqui-sition of large U.S. corporations in the 1960s: The other contested terrain. Am.

Sociol. Rev. 60, 469–499.

Pfeffer, J., Salancik, G.R., 1978. The External Control of Organizations: A ResourceDependence Perspective. Harper & Row, New York, NY.

Podolny, J.M., 1993. A status-based model of market competition. Am. J. Sociol. 98,829–872.

cial Ne

P

P

P

P

P

P

R

S

J. Owen-Smith et al. / So

odolny, J.M., 2001. Networks as the pipes and prisms of the market. Am. J. Sociol.107, 33–60.

odolny, J.M., Baron, J.N., 1997. Resources and relationships: social networks andmobility in the workplace. Am. Sociol. Rev. 62, 673–693.

odolny, J.M., Toby, S.E., Hannan, M.T., 1996. networks, knowledge, and niches: com-petition in the worldwide semiconductor industry, 1984–1991. Am. J. Sociol.102, 659–689.

ollock, T.G., Rindova, V.P., Maggitti, P.G., 2008. Market watch: Information andavailability cascades in the U.S. IPO Market. Acad. Manage. J. 51, 335–358.

owell, W.W., Koput, K.W., Bowie, J.I., Smith-Doerr, L., 2002. The spatial clustering ofscience and capital: Accounting for biotech firm-venture capital relationships.Regional Stud. 36, 291–305.

ollock, T.G., Rindova, V.P., 2003. Media legitimation effects in the market for initial

public offerings. Acad. Manage. J. 46, 631–642.

itter, J., Welch, I., 2002. A Review of IPO activity, pricing, and allocations. J. Financ.57, 1795–1828.

mith, N., Menon, T., Thompson, L., 2012. Status differences in the cognitive activa-tion of social networks. Organ. Sci. 23 (1), 67–82.

tworks 41 (2015) 1–17 17

Saxenian, A., 1994. Regional Advantage: Culture and Competition in Silicon Valleyand Route 128. Harvard University Press, Cambridge, MA.

Stross, R.E., 2000. eBoys: The First Inside Account of Venture Capitalists at Work.Crown Business, New York, NY.

Stuart, T.E., Hoang, H., Hybels, R.C., 1999. Interorganizational endorsements and theperformance of entrepreneurial ventures. Admin. Sci. Q. 44, 315–349.

Useem, M., 1982. Classwide rationality in the politics of managers and directorsof large corporations in the United States and Great Britain. Admin. Sci. Q. 27,199–226.

Useem, M., 1984. The Inner Circle: Large Corporations and the Rise of BusinessPolitical Activity in the U.S. and U.K. Oxford University Press, New York.

Uzzi, B., 1996. The sources and consequences of embeddedness for the eco-nomic performance of organizations: the network effect. Am. Sociol. Rev. 61,

674–698.

Zhao, X., (Doctoral dissertation) 2005. What causes an initial public offering to beunsuccessful? An empirical analysis. Mississippi State University.

Zuckerman, E.W., 1999. The categorical imperative: securities analysts and the ille-gitimacy discount. Am. J. Sociol. 104, 1398–1438.